Position

This file holds what is known, what is not known, and what is not claimed. It also holds a proposal for iterative deepening for reading.

Known

  • A map, outline or topic given before the text helps memory and comprehension (organizers small, maps larger).
  • Telling the reader what matters raises learning of that part and lowers learning of the rest.
  • Examples before problems, and alternating concept and procedure lessons, beat their reverses.
  • Interleaving practice across units beats blocking.
  • In hypertext, a related-next order helps readers with low knowledge.
  • A self-test, not a feeling of understanding, is a sound signal to stop.

Not known

  • No trial compares breadth-first and depth-first reading of the same hierarchical material.
  • Elaboration theory, zoom-lens and spiral sequencing have no positive controlled test.
  • Stop thresholds (minutes, number of descents, number of recalls) come from word-pair laboratory tasks.
  • Whether goal-bounded depth pays off when the later test asks outside the goal.
  • Whether instruction first or problem first wins for a given course: the moderator is element interactivity, and no study of a specific course exists.
  • Depth over breadth (Schwartz 2009) is correlational and about school science.

What is not claimed

  • "Nobody has done this" is not claimed; the review did not find a trial, which is weaker.
  • The proposal below is built from neighbouring results, not from a test of itself.

Proposal: iterative deepening for reading

Grade: convention, untested. Thresholds are PROVISIONAL.

Definitions: - Unit: the chapter or section studied now. - Depth 1: the one-line summary of each notion (hub row, glossary line). - Depth 2: the notion's own page: definition and example. - Depth 3: proofs, derivations, and every link that leaves the unit. - Task set of a unit: notions the unit's exercises or goals require, plus notions marked as must-know, plus their ancestors on the unit's map.

  1. Subject map, once; no link opened.
  2. The next unit whose prerequisites are done.
  3. Depth 1 for all notions; no link opened.
  4. Read the exercise requirements and list the task set; try the first question for 5 min (PROVISIONAL).
  5. Depth 2 for the task set only, in map order; after each, explain it in two lines from memory.
  6. A check failed? Depth 3 on that link only, then check again.
  7. A worked example, then a paired exercise; stalled for 10 min (PROVISIONAL), unfold one step.
  8. The unit is done when a self-test passes unaided, and again a day later.

Stop rule, per notion: 1. Depth 1 always. 2. Depth 2 if and only if the notion is in the task set. 3. Depth 3 only when a failed check traces to that link, or the proof is marked must-know. 4. Go back up when any holds: the check passes from memory, or two descents in a row changed no answer (PROVISIONAL). Also go back up when the link leaves the unit (log it as a black box: what it gives, not how). 5. The unit is done when the self-test passes, and again a day later.

Evidence per step:

Step Evidence Grade
1, 3 context before; maps; prequestions measured
2 prerequisite order supported with caveat
4 relevance instructions; agenda; first-step test measured
5 coherent order; explanation test measured
6 self-explanation of proofs measured
7 example then problem; instruction first for interacting elements measured; supported with caveat
8 criterion level; overconfidence measured
The combination none convention, untested

Fit of the three common habits:

Habit Fits Fails
Overview first steps 1 and 3 read as a list to remember, it is rereading
Depth-first through links step 5 inside the task set unbounded: no stop test, depth spent where nothing asks for it
Breadth-first only steps 3 and 4 as the whole method: skimming gives moderate comprehension and must hand over to steps 5 to 8

Changes to a corpus that would support the procedure: 1. A column showing which exercises need each notion (generated, not typed): makes the task set visible at step 3. 2. A section for links beyond the unit's level, one line each: what it gives, when it would matter. 3. One page on how to read a unit (the steps and the stop rule), linked from each unit. 4. A short folded plan above each must-know proof. 5. An interleaved review step after every two or three units (PROVISIONAL).

Practice and the evidence

  • Practice runs ahead of the evidence: BLUF and the Pyramid Principle have wide use, and no controlled comparison with narrative order for working readers was found.
  • Practice ignores the evidence: the inverted pyramid survives while its two direct tests disagree. Narrative order did better for Thai news readers (Nildam 1998); importance order did marginally better on comprehension, with no difference in memory (Sternadori 2008).
  • Reader state limits the rule: front-loading fits readers who scan for a fact (Nielsen 2006), and explanation first fits bad news that readers do not expect (Jansen 2010). Message valence was manipulated, and surprise is a proposed, untested mechanism.
  • ISO 2023 agrees on order and extends it. It puts importance first at the beginning (5.2.2 a, a convention with no evidence cited in the preview) and builds new information on known information. It adds relevance and evaluation with readers.